multi-level zero inflated models

[James Rao]

Dear All,

I would like to estimate a multi-level zero inflated poisson (ZIP) and zero inflated negative bionomial (ZINB) where the coefficients are allowed to vary across different levels of the count outcome. Which is the most appropriate Stata command for such a model. I have taken a look at the gllamm and the meglm but am not sure whether they would be appropriate.

James

[Maarten Buis]

gllamm might work, meglm won't estimate zero-inflated models. Alternatively, you might get gsem to estimate these models.

[James Rao]

Thanks Maertens,

I have taken a closer look at the gllamm command and I worry that it may not work for the multi-level ZIP or ZINB ......... or rather I am not sure how I would implement it in the case of zero inflated models. Remember there are two stages in a ZIP or ZINB model - the inflation/logistic component and the poisson/negative binomial part. I have seen reference to the "proc lnmixed" command in SAS. See th elink below:

http://www.ats.ucla.edu/stat/sas/faq/zip_nlmixed.htm


Any idea of what could be an equivalent command in Stata?

James

[James Rao]

Dear Maarten,

Returning the gllamm command, kindly advise if it would apply to my case.

I am estimating a zero inflated model (ZIP and ZINB) - demand for artificial insemination (AI) technology by dairy farmers. My dependent variable is a count variable - frequency of AI use over a period of time. My data is of course over-dispersed and that is why I am using ZIP/ZINB. One of the reviewers of my article has suggested the use of a more flexible model where the parameters (coefficients) are unrestricted i.e., they are allowed to vary according to the level of outcome variable. Do you think the generalized linear model (gllamm) would fit my analyses?

I would appreciate your insight.

James

[Maarten Buis]

I suspect (but I am not sure, as I don't know gllamm too well, and I don't understand the exact model you whish to estimate) that in theory you could make a model like that work with gllamm. Finding the right syntax is often a real challenge, and even when you get that right there is no guarantee that the model will converge. Good luck.

[James Rao]

Thanks Maarten. I will try and keep you updated!

[susielee]

Hi James, I am also trying to run a similar model. Have you been able to work through stata gllamm?